fix(media-understanding): place Qwen/DashScope image prompts in user content

XuZehan-iCenter · XuZehan-iCenter · commit 97281d8b73a3 · 2026-06-13T18:25:55.000Z
DashScope vision models (qwen3.7-max, qwen3.7-plus) return 400 "Unexpected item type in content" when the image request sends the prompt in a system message and only image blocks in the user message. Add DashScope endpoint detection to shouldPlaceImagePromptInUserContent so the prompt text is placed in the user content alongside the image, matching how GitHub Copilot and OpenRouter routes already work. Fixes #92688
diff --git a/src/media-understanding/image.test.ts b/src/media-understanding/image.test.ts
@@ -884,6 +884,59 @@ describe("describeImageWithModel", () => {
     ]);
   });
 
+  it("places DashScope/Qwen image prompts in user content before images", async () => {
+    discoverModelsMock.mockReturnValue({
+      find: vi.fn(() => ({
+        api: "openai-completions",
+        provider: "qwen",
+        id: "qwen3.7-max",
+        input: ["text", "image"],
+        baseUrl: "https://dashscope.aliyuncs.com/compatible-mode/v1",
+      })),
+    });
+    completeMock.mockResolvedValue({
+      role: "assistant",
+      api: "openai-completions",
+      provider: "qwen",
+      model: "qwen3.7-max",
+      stopReason: "stop",
+      timestamp: Date.now(),
+      content: [{ type: "text", text: "qwen dashscope ok" }],
+    });
+
+    const result = await describeImageWithModel({
+      cfg: {},
+      agentDir: "/tmp/openclaw-agent",
+      provider: "qwen",
+      model: "qwen3.7-max",
+      buffer: Buffer.from("png-bytes"),
+      fileName: "image.png",
+      mime: "image/png",
+      prompt: "Describe the image.",
+      timeoutMs: 1000,
+    });
+
+    expect(result).toEqual({
+      text: "qwen dashscope ok",
+      model: "qwen3.7-max",
+    });
+    const firstCall = requireFirstMockCall(completeMock, "DashScope image completion");
+    const [, context] = firstCall;
+    expect(context.systemPrompt).toBeUndefined();
+    const userMessage = context.messages[0];
+    if (!userMessage) {
+      throw new Error("expected DashScope image completion user message");
+    }
+    expect(userMessage.content).toEqual([
+      { type: "text", text: "Describe the image." },
+      {
+        type: "image",
+        data: Buffer.from("png-bytes").toString("base64"),
+        mimeType: "image/png",
+      },
+    ]);
+  });
+
   it.each([
     {
       name: "direct OpenAI Responses baseUrl",
diff --git a/src/media-understanding/image.ts b/src/media-understanding/image.ts
@@ -281,6 +281,17 @@ function buildImageContext(
   };
 }
 
+function isDashScopeEndpoint(model: Model): boolean {
+  const provider = model.provider?.toLowerCase() ?? "";
+  const baseUrl = model.baseUrl?.toLowerCase() ?? "";
+  return (
+    provider.includes("dashscope") ||
+    provider === "qwen" ||
+    provider === "qwen-dashscope" ||
+    baseUrl.includes("dashscope.aliyuncs.com")
+  );
+}
+
 function shouldPlaceImagePromptInUserContent(model: Model): boolean {
   // GitHub Copilot models (including Gemini 3.1 Pro Preview) require the
   // prompt text to be in the user message alongside the image. Placing it
@@ -289,6 +300,12 @@ function shouldPlaceImagePromptInUserContent(model: Model): boolean {
   if (model.provider === "github-copilot") {
     return true;
   }
+  // DashScope/Qwen vision models return 400 "Unexpected item type in content"
+  // when the user message contains only image blocks with no text. The prompt
+  // must be placed in the user content alongside the image.
+  if (isDashScopeEndpoint(model)) {
+    return true;
+  }
   const capabilities = resolveProviderRequestCapabilities({
     provider: model.provider,
     api: model.api,
